Mixing times for a constrained Ising process on the torus at low density (Q2412665)

The paper considers a kinetically constrained Ising process (KCIP) associated with a given graph \(G=(V,E)\) and density parameter \(p\). This is an interacting particle system with state space \(\{0,1\}^G\) with the condition that there is at least one particle in the system. There is a reversible Markov chain \(X_t\) on the \(\{0,1\}\)-labellings of \(G\), and a vertex can change its state if at least one neighbour is in state 1 depending on the vertex and the density parameter \(p\). For a positive integer \(L\in \mathbb N\), let \(\Lambda (L,d)\) denote the \(d\)-dimensional torus with \(n=L^d\) points; the authors study the KCIP on a sequence of graphs \(\{\Lambda (L,d)\}_{L\in \mathbb N}\) with density \(p=p_n=\frac{c}{n}\) for some fixed constant \(c<\infty \) and fixed dimension \(d\geq 3\). The authors prove that the mixing time of the KCIP on \(\Lambda (L,d)\) satisfies the inequality \[ C_1n^3\leq \tau _{\mathrm{mix}}\leq C_2n^3\log (n) \] for some constants \(C_1\), \(C_2\), depending on \(c\) and \(d\), but independent of \(n\). The low-density regime \(p=c/| G|\) is subject to Aldous' conjecture, the results provide a counterexample to it and suggest a natural modification. The model was introduced by physicists for glass transition, and recently as an algorithm for data storage in computer networks.